/* PSPP - a program for statistical analysis.
- Copyright (C) 2009 Free Software Foundation, Inc.
+ Copyright (C) 2009, 2010, 2011 Free Software Foundation, Inc.
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
#include <config.h>
-#include <stdio.h>
+#include "math/categoricals.h"
+#include "math/interaction.h"
-#include "categoricals.h"
+#include <stdio.h>
-#include <gl/xalloc.h>
-#include <data/variable.h>
-#include <data/case.h>
-#include <data/value.h>
-#include <libpspp/hmap.h>
-#include <libpspp/pool.h>
+#include "data/case.h"
+#include "data/value.h"
+#include "data/variable.h"
+#include "libpspp/array.h"
+#include "libpspp/hmap.h"
+#include "libpspp/pool.h"
+#include "libpspp/str.h"
-#include <libpspp/str.h>
+#include "gl/xalloc.h"
struct value_node
{
struct hmap_node node; /* Node in hash map. */
- union value value; /* The value being labeled. */
+ struct ccase *ccase;
double cc; /* The total of the weights of cases with this value */
+
+ void *user_data; /* A pointer to data which the caller can store stuff */
+
int subscript; /* A zero based integer, unique within the variable.
Can be used as an index into an array */
};
-
struct var_params
{
/* A map indexed by a union values */
struct hmap map;
- const struct variable *var;
+ const struct interaction *iact;
- int base_subscript;
+ int base_subscript_short;
+ int base_subscript_long;
/* The number of distinct values of this variable */
int n_cats;
};
+/* Comparison function to sort the reverse_value_map in ascending order */
+static int
+compare_value_node (const void *vn1_, const void *vn2_, const void *aux)
+{
+ const struct value_node * const *vn1 = vn1_;
+ const struct value_node * const *vn2 = vn2_;
+ const struct var_params *vp = aux;
+
+ return interaction_case_cmp_3way (vp->iact, (*vn1)->ccase, (*vn2)->ccase);
+}
+
+
struct categoricals
{
/* The weight variable */
const struct variable *wv;
-
/* An array of var_params */
struct var_params *vp;
In the absence of missing values, this will be equal to N_VP */
size_t n_vars;
- /* A map to enable the lookup of variables indexed by subscript */
- int *reverse_variable_map;
+ /* A map to enable the lookup of variables indexed by subscript.
+ This map considers only the N - 1 of the N variables.
+ */
+ int *reverse_variable_map_short;
+
+ /* Like the above, but uses all N variables */
+ int *reverse_variable_map_long;
size_t n_cats_total;
/* Missing values to be excluded */
enum mv_class exclude;
+
+ /* Function to be called on each update */
+ update_func *update;
+
+ /* Function specified by the caller to create user_data */
+ user_data_create_func *user_data_create;
+
+ /* Auxilliary data to be passed to update and user_data_create_func*/
+ void *aux1;
+ void *aux2;
};
int x;
printf ("\n%s (%d) CC=%g n_cats=%d:\n",
- var_get_name (vp->var), vp->base_subscript, vp->cc, vp->n_cats);
+ var_get_name (vp->var), vp->base_subscript_long, vp->cc, vp->n_cats);
printf ("Reverse map\n");
for (x = 0 ; x < vp->n_cats; ++x)
printf ("Number of non-empty categorical variables: %d\n", cat->n_vars);
printf ("Total number of categories: %d\n", cat->n_cats_total);
- printf ("\nReverse variable map:\n");
+ printf ("\nReverse variable map (short):\n");
+ for (v = 0 ; v < cat->n_cats_total - cat->n_vars; ++v)
+ printf ("%d ", cat->reverse_variable_map_short[v]);
+ printf ("\nReverse variable map (long):\n");
for (v = 0 ; v < cat->n_cats_total; ++v)
- printf ("%d ", cat->reverse_variable_map[v]);
+ printf ("%d ", cat->reverse_variable_map_long[v]);
+
printf ("\n");
}
#endif
-
static struct value_node *
-lookup_value (const struct hmap *map, const struct variable *var, const union value *val)
+lookup_case (const struct hmap *map, const struct interaction *iact, const struct ccase *c)
{
- struct value_node *foo;
- unsigned int width = var_get_width (var);
- size_t hash = value_hash (val, width, 0);
+ struct value_node *nn;
+ size_t hash = interaction_case_hash (iact, c);
- HMAP_FOR_EACH_WITH_HASH (foo, struct value_node, node, hash, map)
+ HMAP_FOR_EACH_WITH_HASH (nn, struct value_node, node, hash, map)
{
- if (value_equal (val, &foo->value, width))
+ if (interaction_case_equal (iact, c, nn->ccase))
break;
fprintf (stderr, "Warning: Hash table collision\n");
}
- return foo;
+ return nn;
}
-
struct categoricals *
-categoricals_create (const struct variable *const *v, size_t n_vars,
- const struct variable *wv, enum mv_class exclude)
+categoricals_create (const struct interaction **inter, size_t n_inter,
+ const struct variable *wv, enum mv_class exclude,
+ user_data_create_func *udf,
+ update_func *update, void *aux1, void *aux2
+ )
{
size_t i;
struct categoricals *cat = xmalloc (sizeof *cat);
- cat->n_vp = n_vars;
+ cat->n_vp = n_inter;
cat->wv = wv;
cat->n_cats_total = 0;
cat->n_vars = 0;
- cat->reverse_variable_map = NULL;
+ cat->reverse_variable_map_short = NULL;
+ cat->reverse_variable_map_long = NULL;
cat->pool = pool_create ();
cat->exclude = exclude;
+ cat->update = update;
+ cat->user_data_create = udf;
+
+ cat->aux1 = aux1;
+ cat->aux2 = aux2;
+
cat->vp = pool_calloc (cat->pool, cat->n_vp, sizeof *cat->vp);
for (i = 0 ; i < cat->n_vp; ++i)
{
hmap_init (&cat->vp[i].map);
- cat->vp[i].var = v[i];
+ cat->vp[i].iact = inter[i];
}
return cat;
const double weight = cat->wv ? case_data (c, cat->wv)->f : 1.0;
- assert (NULL == cat->reverse_variable_map);
+ assert (NULL == cat->reverse_variable_map_short);
+ assert (NULL == cat->reverse_variable_map_long);
for (i = 0 ; i < cat->n_vp; ++i)
{
- const struct variable *var = cat->vp[i].var;
- unsigned int width = var_get_width (var);
- const union value *val = case_data (c, var);
+ const struct interaction *iact = cat->vp[i].iact;
size_t hash;
struct value_node *node ;
- if ( var_is_value_missing (var, val, cat->exclude))
+ if ( interaction_case_is_missing (iact, c, cat->exclude))
continue;
- hash = value_hash (val, width, 0);
- node = lookup_value (&cat->vp[i].map, var, val);
+ hash = interaction_case_hash (iact, c);
+ node = lookup_case (&cat->vp[i].map, iact, c);
if ( NULL == node)
{
node = pool_malloc (cat->pool, sizeof *node);
- value_init (&node->value, width);
- value_copy (&node->value, val, width);
+ node->ccase = case_ref (c);
node->cc = 0.0;
hmap_insert (&cat->vp[i].map, &node->node, hash);
cat->n_vars++;
node->subscript = cat->vp[i].n_cats++ ;
+
+ if (cat->user_data_create)
+ node->user_data = cat->user_data_create (cat->aux1, cat->aux2);
}
node->cc += weight;
cat->vp[i].cc += weight;
+
+ if (cat->update)
+ cat->update (node->user_data, cat->exclude, cat->wv, NULL, c, cat->aux1, cat->aux2);
}
}
}
-/* This function must be called *before* any call to categoricals_get_*_by subscript an
+/* This function must be called *before* any call to categoricals_get_*_by subscript and
*after* all calls to categoricals_update */
void
-categoricals_done (struct categoricals *cat)
+categoricals_done (const struct categoricals *cat_)
{
/* Implementation Note: Whilst this function is O(n) in cat->n_cats_total, in most
uses it will be more efficient that using a tree based structure, since it
1 call of O(n) + 10^9 calls of O(1) is better than 10^9 calls of O(log n).
*/
+ struct categoricals *cat = CONST_CAST (struct categoricals *, cat_);
int v;
- int idx = 0;
- cat->reverse_variable_map = pool_calloc (cat->pool,
- cat->n_cats_total,
- sizeof *cat->reverse_variable_map);
+ int idx_short = 0;
+ int idx_long = 0;
+ cat->reverse_variable_map_short = pool_calloc (cat->pool,
+ cat->n_cats_total - cat->n_vars,
+ sizeof *cat->reverse_variable_map_short);
+
+ cat->reverse_variable_map_long = pool_calloc (cat->pool,
+ cat->n_cats_total,
+ sizeof *cat->reverse_variable_map_long);
for (v = 0 ; v < cat->n_vp; ++v)
{
vp->reverse_value_map = pool_calloc (cat->pool, n_cats_total, sizeof *vp->reverse_value_map);
- vp->base_subscript = idx;
+ vp->base_subscript_short = idx_short;
+ vp->base_subscript_long = idx_long;
for (node = hmap_first (&vp->map); node; node = hmap_next (&vp->map, node))
{
vp->reverse_value_map[vn->subscript] = vn;
}
- /* Populate the reverse variable map.
- */
+ /* For some purposes (eg CONTRASTS in ONEWAY) the values need to be sorted */
+ sort (vp->reverse_value_map, vp->n_cats, sizeof (const struct value_node *),
+ compare_value_node, vp);
+
+ /* Populate the reverse variable maps. */
+ for (i = 0; i < vp->n_cats - 1; ++i)
+ cat->reverse_variable_map_short[idx_short++] = v;
+
for (i = 0; i < vp->n_cats; ++i)
- cat->reverse_variable_map[idx++] = v;
+ cat->reverse_variable_map_long[idx_long++] = v;
}
assert (cat->n_vars <= cat->n_vp);
static int
-reverse_variable_lookup (const struct categoricals *cat, int subscript)
+reverse_variable_lookup_short (const struct categoricals *cat, int subscript)
{
- assert (cat->reverse_variable_map);
+ assert (cat->reverse_variable_map_short);
+ assert (subscript >= 0);
+ assert (subscript < cat->n_cats_total - cat->n_vars);
+
+ return cat->reverse_variable_map_short[subscript];
+}
+
+static int
+reverse_variable_lookup_long (const struct categoricals *cat, int subscript)
+{
+ assert (cat->reverse_variable_map_long);
assert (subscript >= 0);
assert (subscript < cat->n_cats_total);
- return cat->reverse_variable_map[subscript];
+ return cat->reverse_variable_map_long[subscript];
}
+
/* Return the categorical variable corresponding to SUBSCRIPT */
const struct variable *
categoricals_get_variable_by_subscript (const struct categoricals *cat, int subscript)
{
- int index = reverse_variable_lookup (cat, subscript);
+ int index = reverse_variable_lookup_short (cat, subscript);
+
+ return cat->vp[index].iact->vars[0];
+}
+
+/* Return the interaction corresponding to SUBSCRIPT */
+const struct interaction *
+categoricals_get_interaction_by_subscript (const struct categoricals *cat, int subscript)
+{
+ int index = reverse_variable_lookup_short (cat, subscript);
- return cat->vp[index].var;
+ return cat->vp[index].iact;
}
-/* Return the value corresponding to SUBSCRIPT */
-const union value *
-categoricals_get_value_by_subscript (const struct categoricals *cat, int subscript)
+
+/* Return the case corresponding to SUBSCRIPT */
+static const struct ccase *
+categoricals_get_case_by_subscript (const struct categoricals *cat, int subscript)
{
- int vindex = reverse_variable_lookup (cat, subscript);
+ int vindex = reverse_variable_lookup_short (cat, subscript);
const struct var_params *vp = &cat->vp[vindex];
- const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript];
+ const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript_short];
- return &vn->value;
+ return vn->ccase;
}
double
categoricals_get_weight_by_subscript (const struct categoricals *cat, int subscript)
{
- int vindex = reverse_variable_lookup (cat, subscript);
+ int vindex = reverse_variable_lookup_short (cat, subscript);
const struct var_params *vp = &cat->vp[vindex];
return vp->cc;
double
categoricals_get_sum_by_subscript (const struct categoricals *cat, int subscript)
{
- int vindex = reverse_variable_lookup (cat, subscript);
+ int vindex = reverse_variable_lookup_short (cat, subscript);
const struct var_params *vp = &cat->vp[vindex];
- const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript];
+ const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript_short];
return vn->cc;
}
categoricals_get_binary_by_subscript (const struct categoricals *cat, int subscript,
const struct ccase *c)
{
- const struct variable *var = categoricals_get_variable_by_subscript (cat, subscript);
- int width = var_get_width (var);
+ const struct interaction *iact = categoricals_get_interaction_by_subscript (cat, subscript);
- const union value *val = case_data (c, var);
+ const struct ccase *c2 = categoricals_get_case_by_subscript (cat, subscript);
- return value_equal (val, categoricals_get_value_by_subscript (cat, subscript), width);
+ return interaction_case_equal (iact, c, c2);
}
{
return cat->n_vars;
}
+
+
+/* Return a case containing the set of values corresponding to SUBSCRIPT */
+const struct ccase *
+categoricals_get_case_by_category (const struct categoricals *cat, int subscript)
+{
+ int vindex = reverse_variable_lookup_long (cat, subscript);
+ const struct var_params *vp = &cat->vp[vindex];
+ const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript_long];
+
+ return vn->ccase;
+}
+
+
+void *
+categoricals_get_user_data_by_category (const struct categoricals *cat, int subscript)
+{
+ int vindex = reverse_variable_lookup_long (cat, subscript);
+ const struct var_params *vp = &cat->vp[vindex];
+
+ const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript_long];
+ return vn->user_data;
+}